Research On Collision Detection For 3D Simulation Of Service Robot

The distance returned by three-dimensional simulation of service robot is very important for realizing collision avoidance between robot and environment models as well as collision avoidance between robot joints. It is the foundation of carrying out a specific mission or doing path planning. A complete service robot simulation system must have the ability to do real-time collision detection, including collision detection between robot and environment models and self-detection. The three-dimensional simulation environment is uncertain. Robot has to achieve various achievements according to customers'requirements. As a result, the collision detection system of service robot should have good flexibility and robustness, high detection efficiency and the ability to handle different complex environment. It also should have good collision detection accuracy at the same time. At present, however, most of the collision detection algorithms are only for convex models. There are seldom algorithms for non-convex models. When the number of objects involved in collision detection increase, the efficiency will be obviously decreased, making collision detection becomes an obstacle for service robot simulation.This paper is based on the project"Research & Development of Intelligent Robots and Programming Platform", which is cooperated by Shanghai Jiao Tong University and Yaskawa electric corp. With a comprehensive understanding of all kinds of collision detection algorithms, combining analysis of characteristics in three-dimensional simulation of service robot, we design a set of effective strategy to realize collision detection: Using V-Clip, which has a great detection efficiency, to meet the real-time and algorithm robustness requirement. To solve the problem that non-convex models cannot be directly applied in collision detection algorithm, we design a method to execute decomposition of non-convex models into convex ones to build collision detection units, basing on the data structure of VRML models in simulation. It is completely done by program. An efficient collision detection scheme is designed to provide distance information for robot path planning. To solve the problem that when the number of objects involved in collision detection increase, the efficiency will obviously decline, we use a bounding box technology to achieve coarse detection and pure detection. Automatic division of detection area method is designed to do collision detection in complex environment. A simple friendly user interface is designed to realize all the functions.Through simulation test, this scheme shows high collision detection accuracy, good real-time and robustness. With all the possible exceptions handled, it also possesses a good stability. The user interface is simple to use. Most processes are handled by program automatically, making the system flexible for other applications, which is very suitable for simulation of service robot.